Overall plan: creation of a trio of robots that will intelligently play hide and seek, tagging each other and learning tactics. In outline format, our project consists of...
Introduction
Problem: Getting robots to hide and seek intelligently
Objective: End up with working hiders and seekers
Relation to earlier work: This will combine many aspects of other works covered in our bibliography particularly including the Monte Carlo localization paper and prior work such as described in the Dervish paper
Recent work on this topic: To the best of our knowledge, there has been no attempt to have robots play hide and seek in the fasion we plan on having them do.
Background
Difficulties: We must integrate movement, localization, strategy, vision recognition, wireless communication, and dynamic environments to make optimal robots.
Others in the area: None that we know of (yet)
Motivating features: Essentially those covered in the difficulties with an emphasis on the basics first and higher level processes later.
Scope: The robots will only seek the other robots and will also have some distinctive marking(s) to make the vision more feasible. Advanced strategies will only come after working basics.
Approach
Key design decisions: First, the plan to move from a basic robot that moves up to a fully functional hider and/or seeker is a key decision. This causes us to have modular programming methods so as to easily add or change the various strategies being used by the different robots. The choice of the evolution platform as part of this is also key because of the modularity of its equipment and power of its onboard computing.
Design choice: The platform was chosen in equal parts because of the modularity, the processing power, and the fact that the robotics class already has them. A blimp-based hiding and seeking set of robots, for example, was not pursued because of the amount of time that would have been spent on the robot engineering and weight restriction issues alone.
Difficulties of the design: Not too many. We are somewhat limited in the environment of the robot because of the wheels and the sensativity of (our) laptops to weather and such, but these limits are acceptable. Doing the same seeking in an arbitrary environment rather than primarily white hallways also drastically increases the difficulty of detecting the other robots, especially if they are hidden.
Solutions to difficulties: All are acceptable for now; some sort of change to the evolution's movement system and a method of protecting the laptops would be required if we were planning on extending the robot's operational environment to arbitrary locations.
Progress and Performance
As mentioned below and shown in the picture section, we have our robot moving and beginning to have self-guided wandering and obstacle avoidance with some combination of sonar, IR, and camera senseing.
See the Pictures section of the sidebar for pictures.
Perspective
So far most of the work we've done with statistics and analysis has been in the testing and calibration of the movement uncertainty.
To expand this, we are currently working on implementing Monte Carlo style localization that takes into consideration the uncertainty in the movement of the robot.
In conjunction with this, we are working on wandering that is meant to take the robot eventually to places it hasn't been before as a prelude to the full hiding and seeking.
References
See the Bibliography section of the sidebar for details thus far.
More Specific Plan (original; subject to change)
Get the robots moving (done)
Create object avoidance (in progress)
Create mapping/localization procedures (in progress)
Create vision algorithm to recognize other robots (red vision recognition; uncertainty in robot color makes this unfinished for now).
Create 'tagging' system to change hiders and seekers
Create advanced/statistical/effective hiding/seeking
Further refinement of algorithms, possibly including:
Machine Learning analysis of results
Genetic refinement of algorithm (on a simulator)